It may be that you don't have effects at the level of an individual session due to a lack of power - there could be an effect, but not significant unless you look at it at a group level. How you choose the location of your ROI depends on your question, but it would be unusual to do it on a session-by-session basis. Consider choosing the location based on activations in the group level second-level contrast. Or anatomically if that is more relevant to your question...
P.
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
> On Behalf Of Sevel,Landrew S
> Sent: 28 May 2013 18:22
> To: [log in to unmask]
> Subject: Re: [SPM] DCM-related questions
>
> Thanks for your reply.
>
> I see that it would be preferable to include all sessions and all
> subjects. If activation is not show in one session though, it would not
> be possible to make an an eigenvariate from the contrast showing your
> effects of interest (even if you used an incredibly generous p-value
> the resulting time series would negatively impact resulting model fit
> and interpretation). If you were to throw out one session of three that
> were going to be averaged, the affected subject's parameters would be
> based upon a different amount information than others. To completely
> remove this subject would at the same time lose two sessions of
> potentially relevant information. Is there a preferable course of
> action?
>
> Best,
>
> Drew
> ________________________________________
> From: Zeidman, Peter [[log in to unmask]]
> Sent: Tuesday, May 28, 2013 6:24 AM
> To: Sevel,Landrew S; [log in to unmask]
> Subject: RE: DCM-related questions
>
> Hi Drew,
>
> > If we were to concatenate all three sessions, would the scale of each
> time series be in the same metric? If this is not the case, are there
> any sets I can take to correct this issue?
>
> It's best not to concatenate, but if you need to, you'll need to add an
> extra regressor for all but one session in your GLM. I.e. given three
> sessions, you'll have a column marking every scan from session 1 and a
> column for every scan from session 2. These will model the mean for
> each session. (You shouldn't have a regressor for the last session, or
> your model will be over-specified).
>
> > In the event that we were not to concatenate sessions but to average
> parameters across models after model selection for parameter comparison
> between groups, are there any guidelines in the event that one or two
> of three sessions does not show significant activation in our VOIs?I'm
> aware that it is standard to remove single subjects from analysis if
> they do not obtain significant activation in VOIs, but what about
> single sessions?
>
> In general I would include all sessions and subjects, not just those
> showing the effects at the individual level.
>
> > Lastly on parameter interpretation. For the connection A-->B,
> increases in activity of B correspond to 10% of the activity in A per
> unit time when the parameter is 0.10. Can these values exceed 1.00 and
> if so how could this be interpreted?
>
> Someone else may know better than me - but I believe they can go over
> 1, which would mean a value of 1.1 would mean region B increases by
> 1.1 times the activity in unit A in the context of your modulation.
>
> Best,
> Peter.
>
|